Yes, you can. GCTA-GREML does not assume that the individuals should be unrelated. The reason for excluding close-relatives in Yang et al. (Nat. Genet. 2010 and 2011) is because we do not want our estimates to be confounded with some possible shared environment effects and the effects of some possible causal variants that are not tagged by the SNPs but captured by pedigree information. If you are interested in the variance explained by a subset of SNPs in family data, you could fit the genetic relationship matrix (GRM) estimated from these SNPs along with a matrix of pedigree structure using the option --mgrm when running the REML analysis (--reml). Alternatively, we could fit the GRM of the subset of SNPs together with another GRM estimated from the SNPs in the rest of the genome.